Connect with us

Hi, what are you looking for?

Tuesday, Feb 11, 2025
Mugglehead Investment Magazine
Alternative investment news based in Vancouver, B.C.

AI and Autonomy

Zurich’s Lakera raises millions to protect LLMs from malicious users

Punching the right combination of words into these programs can trick them into revealing things they shouldn’t

Switzerland's Lakera raises millions to help protect LLMs from malicious prompts
This startup has created "Gandalf," a game where users try to trick an LLM into revealing secret passwords. Photo credit: Lakera

Europe’s venture capital firm Atomico has helped the Swiss AI developer Lakera raise over C$27 million to propel its large language model (LLM) security tech.

The startup, founded in 2021, revealed that its Series A funding round had been completed on Jul. 24. New York’s Citigroup Inc (NYSE: C), Dropbox Inc (Nasdaq: DBX) and Lakera stakeholders like Switzerland’s redalpine helped raise funds from investors too.

“Generative AI introduces new attack methods, including prompt and data poisoning attacks, which can exploit the model to take unintended actions,” Lakera said in a news release.

Users with malicious intent can use certain prompt phrases to trick an LLM into revealing confidential info or make the model behave inappropriately. Lakera was established to solve this problem.

Only AI technology is capable of protecting AI programs from the dangers posed by hackers knowledgeable about AI, the company says.

“Lakera uses proprietary AI to secure AI applications, so our customers stay ahead of continuously evolving threats.”

Former engineers from Meta Platforms Inc (Nasdaq: META) and Google co-founded the tech guru. It has offices in Zurich and San Francisco.

Read more: VERSES AI levels up with global standards for intelligent system interoperability

Read more: Illustrious market research firm Gartner recognizes VERSES AI for its innovation

Lakera creates novel wizard game

Gandalf was invented so that people can learn about AI security in a fun way, Lakera says.

In the game, the user tries to get the LLM wizard to reveal a password. There are eight levels, which get progressively harder as Gandalf tightens his grip on the words.

“Gandalf will upgrade the defences after each successful password guess!” Lakera explained. The data pool generated through interactions with the young virtual sorcerer grows immensely every day.

“Our models continuously learn from large amounts of generative AI interactions what malicious interactions look like,” chief executive David Haber told TechCrunch.

Zurich's Lakera raises millions to protect LLMs from malicious prompts

Gandalf. Image credit: Lakera

Lakera collaborates with Cohere

Gandalf’s creator has collaborated with a Canadian AI company that just raised nearly C$700 million for its data security initiatives. Cohere will benefit from Lakera’s LLM knowledge while developing its own, such as Rerank 3.

“The team has extensive expertise and deep understanding of complex security challenges,” Cohere safety head Seraphina Goldfarb-Tarrant said.

Generative AI developers in various niches within the AI sector could potentially benefit from Lakera’s innovation. VERSES AI Inc. (CBOE: VERS) (OTCQB: VRSSF), developer of the neuroscience and biology-focused AI program Genius, is one of them.

“Cybersecurity is paramount to fostering a thriving ecosystem of intelligent systems,” VERSES says. 

 

VERSES AI is a sponsor of Mugglehead news coverage 

 

Follow Mugglehead on X

Like Mugglehead on Facebook

Follow Rowan Dunne on X

rowan@mugglehead.com

Click to comment

Leave a Reply

Your email address will not be published. Required fields are marked *

You May Also Like

AI and Autonomy

Shared Services Canada has already applied the restriction to mobile devices it manages

AI and Autonomy

Over 3,800 people have signed a petition calling for it to be cancelled

AI and Autonomy

Anthropic, OpenAI and other startups are seeking investments to sustain operations

AI and Autonomy

OpenAI CEO Sam Altman recently said that India could be a leading force in the AI revolution